Accurate Calculation Of The Free Energy In Bio-macromolecular Systems Based On The MM/PBSA Method

There are a large number of biological macromolecules such as proteins,nucleic acids and polysaccharides inside and outside organisms or cells.These biomacromolecules lay the foundation for the life activities through the interactions between themselves or with small ligands.A deep understanding of the interactions among the biomacromolecules and their ligands is of great scientific significance in understanding the underlying mechanisms of life processes,and the calculation of free energy related to biomacromolecules can help predict or analyze the interaction mechanisms and binding states between biomacromolecules and their ligands.Therefore,the accurate calculation of the free energy of macromolecular systems is extremely important in the fields of biophysics and biomedicine,but it is still a very challenging problem.In this paper,we propose an improved free energy calculation method based on molecular mechanics/Poisson-Boltzmann surface area(MM/PBSA)method,and further improve the reliability and accuracy of the calculation method by combining with machine learning methods.In chapter 1,we first introduce the types of interactions in biological macromolecules and their biological significance.Then the basic idea,theoretical basis and specific steps of molecular dynamics simulation are introduced in detail.The main methods of free energy calculation for bio-macromolecular systems and their advantages and disadvantages are discussed.Moreover,several common machine learning methods are briefly introduced.Finally,we also briefly introduce the main research content and significance of this paper.In chapter 2,we introduce the electrostatic shielding effect on the basis of standard MM/PBSA method(i.e.adding exponential damping factor into coulomb interaction energy),and propose a screening MM/PBSA method.The improved MM/PBSA method and standard MM/PBSA method are compared in details in the protein-protein system.Our results show that the Pearson correlation coefficient in the improved MM/PBSA is over 0.70,which is much better than that in the standard MM/PBSA,especially in the Amber14SB field.In addition,we calculate the mean absolute error(MAE)value between the calculated and experimental values and find that the MAE in the improved MM/PBSA is much smaller than that in the standard MM/PBSA.In addition,the influences of dielectric constant of protein and salt concentration in solution on the results are investigated.These results indicate that screening MM/PBSA method has a good performance in predicting the binding free energy of highly charged biomolecules.In chapter 3,combining the screening MM/PBSA method and machine learning methods(linear regression,random forest,support vector machine,and backpropagation neural network),we take protein-ligand interaction as an example to study the computational accuracy of binding free energy.We find that the accuracy of calculation results can be greatly improved by using the polar and non-polar terms of the solvation free energy,electrostatic interaction term and Van der Waals interaction term in MM/PBSA method and the key characteristics of ligand or protein structure(such as ligand charge and molecular weight,protein charge,etc.)as characteristic quantities.In addition,we also find that the performance of random forest is better than that of linear fitting method,but the performance of support vector machine and neural network method is only comparable to that of linear fitting method.In chapter 4,we summarize this paper and look forward to the future works.